Intelligent weld detection platform based on artificial intelligence

Abstract

The present invention discloses an intelligent weld detection platform based on artificial intelligence, which relates to the field of intelligent detection technology, including: performing real-time sensing and acquisition of a target welding area through a sensing device group to generate multiple weld image information; traversing multiple weld image information for evaluation, and extracting a weld feature set according to the evaluation result; annotating the multiple weld image information based on the weld feature set to generate multiple image annotation information; constructing a weld quality detection model, synchronizing multiple image annotation information to the weld quality detection model, and obtaining a weld quality detection data set; performing backtracking verification on the weld quality detection data set, updating the weld quality detection data set according to the verification result, and transmitting the updated result to the weld intelligent detection platform. The present invention solves the technical problem that the prior art cannot realize the dynamic update of the weld quality detection data set, resulting in low detection accuracy, and achieves the technical effect of improving the weld detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to an intelligent weld detection platform based on artificial intelligence. Background Art

[0002] With the continuous advancement of industrial production and manufacturing technology, welding technology has been widely used in many fields. However, with the expansion of production scale and the complexity of process flow, the detection and management of weld quality are facing increasing challenges. Although traditional weld detection technology has achieved a certain degree of automation, there are still some problems that cannot be ignored, which makes it difficult to meet the high efficiency, accuracy and safety requirements of modern industrial production for weld quality. A significant problem with traditional technology is that it is impossible to dynamically update the weld quality detection data set. The detection results often lag behind the actual production process, resulting in limited real-time and effectiveness of quality control. Summary of the invention

[0003] The present application provides an artificial intelligence-based weld intelligent detection platform, which is used to solve the technical problem that the existing technology cannot realize the dynamic update of the weld quality detection data set, resulting in low detection accuracy.

[0004] In view of the above problems, this application provides an artificial intelligence-based weld intelligent detection platform.

[0005] The present application provides an artificial intelligence-based weld intelligent detection platform, the platform comprising:

[0006] A sensing acquisition module, wherein the sensing acquisition module performs real-time sensing acquisition of the target welding area through a sensing device group to generate multiple weld image information; a weld feature set extraction module, wherein the weld feature set extraction module traverses the multiple weld image information for evaluation and extracts a weld feature set according to the evaluation result; a labeling module, wherein the labeling module labels the multiple weld image information based on the weld feature set to generate multiple image labeling information; a weld quality detection module, wherein the weld quality detection module constructs a weld quality detection model, synchronizes the multiple image labeling information to the weld quality detection model, and obtains a weld quality detection data set; a backtracking verification module, wherein the backtracking verification module performs backtracking verification on the weld quality detection data set, updates the weld quality detection data set according to the verification result, and transmits the updated result to the weld intelligent detection platform.

[0007] The technical solution provided in this application has at least the following technical effects or advantages:

[0008] The present application uses a sensor device group to perform real-time sensing acquisition of the target welding area to generate multiple weld image information; traverse multiple weld image information for evaluation, and extract a weld feature set based on the evaluation result; annotate the multiple weld image information based on the weld feature set to generate multiple image annotation information; build a weld quality detection model, synchronize multiple image annotation information to the weld quality detection model, and obtain a weld quality detection data set; perform retrospective verification on the weld quality detection data set, update the weld quality detection data set according to the verification result, and transmit the updated result to the weld intelligent detection platform. The present invention solves the technical problem that the prior art cannot realize the dynamic update of the weld quality detection data set, resulting in low detection accuracy, and achieves the technical effect of improving the weld detection accuracy through real-time sensing acquisition, weld feature extraction, image annotation, retrospective verification and dynamic update. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of the structure of an intelligent weld detection platform based on artificial intelligence provided in an embodiment of the present application;

[0011] Figure 2 A schematic flow chart of an artificial intelligence-based weld intelligent detection method provided in an embodiment of the present application.

[0012] Explanation of the reference numerals: sensing acquisition module 11 , weld feature set extraction module 12 , labeling module 13 , weld quality detection module 14 , backtracking verification module 15 . DETAILED DESCRIPTION

[0013] This application provides an artificial intelligence-based weld intelligent detection platform to solve the technical problem that the existing technology cannot dynamically update the weld quality detection data set, resulting in low detection accuracy. Through real-time sensor acquisition, weld feature extraction, image annotation, backtracking verification and dynamic updating, the technical effect of improving weld detection accuracy is achieved.

[0014] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0015] It should be noted that any variations of the terms "include" and "have" are intended to cover non-exclusive inclusions. For example, a process, method, platform, product or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules that are not explicitly listed or inherent to these processes, methods, products or devices.

[0016] Embodiment 1

[0017] like Figure 1 As shown, the present application provides an artificial intelligence-based weld intelligent detection platform for performing the following Figure 2 The weld intelligent detection method based on artificial intelligence shown in the figure, the platform includes:

[0018] The sensing acquisition module 11 performs real-time sensing acquisition on the target welding area through a sensing device group to generate a plurality of weld image information.

[0019] In the embodiment of the present application, the sensor acquisition module performs real-time sensor acquisition of the entire welding area through the arranged sensor equipment group to obtain multiple weld image information. The sensor group includes laser sensors and infrared sensors.

[0020] Furthermore, the sensor acquisition module 11 in the platform provided in the application embodiment is also used for:

[0021] Retrieve a historical weld sensor information set of a target welding area, wherein the historical weld sensor information set includes sensor light information, sensor coordinate information, and sensor range information; determine a plurality of layout points based on the sensor light information, the sensor coordinate information, and the sensor range information; set a plurality of sensor devices based on the plurality of layout points, connect the plurality of sensor devices according to a plurality of detection sensor requirement information of the target welding area, and generate a sensor topology structure; generate the sensor device group based on the sensor topology structure.

[0022] In an embodiment of the present application, a historical weld sensor information set of the target welding area is retrieved from a historical database, including sensor light information, sensor coordinate information and sensor range information. The sensor light information is obtained by a laser or optical sensor, and the reflection and refraction characteristics of the light are used to identify the optical features of the weld surface; the sensor coordinate information is obtained by a high-precision positioning sensor, which is a GPS, a laser locator or other device that can provide precise coordinates; the sensor range information is obtained by an infrared sensor or an ultrasonic sensor to determine the effective detection area of ​​each sensor. Next, based on the sensor light information, the sensor coordinate information and the sensor range information, data fusion and optimization algorithms are used to determine the layout points of multiple sensor devices. Different types of sensor data are integrated through data fusion algorithms, such as Kalman filtering technology, to eliminate redundant information and contradictory data; the optimal layout point position is calculated through optimization algorithms, such as genetic algorithms or particle swarm optimization algorithms, to ensure full coverage of the welding area and avoid blind spots. After determining the layout points, multiple sensor devices are installed according to these layout points, including laser sensors, infrared sensors, visual sensors and ultrasonic sensors. Laser sensors use laser triangulation methods to detect surface morphology and dimensional changes in welds; infrared sensors detect temperature distribution based on infrared thermal imaging technology; visual sensors use machine vision technology to capture high-definition images of welds; and ultrasonic sensors detect the internal structure of welds based on ultrasonic flaw detection methods.

[0023] Next, these sensor devices are connected according to the multiple detection sensor requirements of the target welding area to generate a sensor topology. The sensor topology design is based on the network topology optimization algorithm to ensure that the data transmission path between each sensor is the shortest and has redundancy to improve the reliability and efficiency of data transmission. Finally, based on the designed sensor topology, a sensor device group is generated.

[0024] Furthermore, the sensor acquisition module 11 in the platform provided in the application embodiment is also used for:

[0025] Start the sensing device group, perform light balance sensing on the target welding area based on the sensing light information to obtain first sensing information; perform position sensing on the target welding area based on the sensing coordinate information to obtain second sensing information; perform multi-angle sensing on the target welding area based on the sensing range information to obtain third sensing information; synchronize the first sensing information, the second sensing information, and the third sensing information to obtain sensing alignment data; and extract the multiple weld image information based on the sensing alignment data.

[0026] In an embodiment of the present application, first, a control system, such as a PLC, an industrial computer or an embedded system, sends a start signal to the sensor device group to start the sensor device group. Next, light balance sensing is performed based on the sensor light information, and the automatic exposure function of the light sensor or camera is used to monitor and adjust the lighting conditions of the welding area. The light sensor measures the ambient light intensity and light source distribution of the welding area. According to the measurement results, the light intensity of the welding area is adjusted by the light source controller, such as turning on or off the auxiliary light source, adjusting the light source angle, etc. The camera uses automatic exposure or preset exposure parameters to shoot the welding area to ensure that the image brightness is moderate and the details are clear. The image data obtained is the first sensing information. Then, position sensing is performed based on the sensing coordinate information, and a laser rangefinder is used to determine the exact position of the welding area. The laser rangefinder emits a laser beam to the welding area and measures the time of reflection back, thereby calculating the distance and position information, forming the second sensing information, and describing the specific position of the welding area in space. Then, multi-angle sensing is performed based on the sensing range information, and multi-angle sensing is achieved by a rotating platform. The rotating platform is controlled to move the camera to each preset sensing angle. At each position, the camera captures an image of the welding area. The above steps are repeated until the entire welding area is covered. The image data from multiple angles obtained is the third sensing information, which provides multi-perspective observation of the welding area.

[0027] After obtaining the first sensor information, the second sensor information and the third sensor information, information synchronization is performed to obtain sensor alignment data. The sensor data collected at different times are aligned using a time synchronization algorithm, and a timestamp is added to each sensor information to ensure their synchronization in time. The sensor alignment data is obtained through this process.

[0028] Finally, the weld image information is extracted based on the sensor alignment data, and the image processing algorithm is used to extract the image information of the weld area from the sensor alignment data. Specifically, the image in the sensor alignment data is preprocessed, such as denoising, and the edge detection algorithm is used to identify the boundary of the weld area, and finally multiple clear and accurate weld image information is extracted from the sensor alignment data.

[0029] Furthermore, the sensor acquisition module 11 in the platform provided in the application embodiment is also used for:

[0030] The sensing range information is divided based on the regional characteristics of the target welding area to obtain multiple sensing sub-areas; the multiple sensing sub-areas are traversed to perform multi-perspective matching to generate multiple sensing angles; the weld is continuously sensed according to the multiple sensing angles to generate multiple perspective image information; the multiple perspective images are fused according to the sensing time series to obtain the third sensing information.

[0031] In an embodiment of the present application, when the sensing range information is divided based on the regional features of the target welding area, the welding area is first divided using image segmentation technology, such as an edge detection method, to obtain multiple sensing sub-areas. The edge detection method uses an operator, such as the Canny operator, to identify significant edges and feature points in the welding area, and divides the entire welding area into multiple small sub-areas. Next, multiple sensing sub-areas are traversed for multi-view matching, and feature matching algorithms in computer vision, such as the SIFT algorithm, are used to extract and match feature points for each sub-area. The SIFT algorithm is used to detect and describe local features in the image, and the optimal sensing angle is determined by comparing the matching of feature points at different angles. Thereafter, the weld is continuously sensed at multiple sensing angles, and a rotating platform or a multi-camera system is used to continuously shoot the weld at multiple angles according to multiple preset sensing angles to obtain multiple perspective image information.

[0032] Finally, the image stitching technology is used to fuse multiple view images according to the sensing time sequence. Specifically, an image registration algorithm, such as the phase correlation method, is used to spatially align images taken at different angles and at different times. Image registration calculates the transformation relationship between images, including translation, rotation, and scaling, by identifying common feature points or feature areas in the image, so as to align the images in the same coordinate system. When the aligned images are fused, the Laplacian pyramid fusion technology is used to construct a Gaussian pyramid for each aligned image. The Gaussian pyramid generates a set of images with decreasing resolution layer by layer through Gaussian blurring and downsampling layer by layer. The Laplacian pyramid is calculated from the Gaussian pyramid. Each layer of the Laplacian pyramid image is a difference map between the corresponding Gaussian pyramid image and its upper level image. On each layer of the Laplacian pyramid, multiple images are fused layer by layer. The fusion method uses the pixel value of the clearest part of each image or the weighted average method. The fused Laplacian pyramid image is reconstructed layer by layer to generate the final fused image and obtain the third sensing information.

[0033] The weld feature set extraction module 12 traverses the plurality of weld image information for evaluation, and extracts a weld feature set according to the evaluation result.

[0034] In the embodiment of the present application, firstly, image preprocessing is performed on multiple weld image information, and Gaussian filtering, mean filtering or median filtering and other techniques are used to remove noise and interference in the image to improve the image quality. Then, edge detection algorithms such as Canny edge detection or Sobel operator are used to identify the edge of the weld, and the morphological characteristics and contour information of the weld are extracted. Next, texture analysis methods such as grayscale co-occurrence matrix are used to extract texture features of the weld image, and texture features such as contrast, entropy, energy and uniformity are calculated. These features can reflect the fine structure and quality of the weld surface.

[0035] In addition, the weld image is processed through morphological processing such as expansion, corrosion, opening and closing operations to enhance the structural features of the weld, remove small artifacts and noise, and make the weld features more obvious. Feature point detection algorithms such as Harris corner detection and SIFT algorithm are used to extract key points in the weld image. Statistical analysis is performed on the extracted edge features, texture features, and morphological features, and statistical quantities such as the mean, variance, and standard deviation of the features are calculated to quantify the distribution and change patterns of the weld features.

[0036] According to the statistical results of the features and the weld quality standards, the features are evaluated and screened. The evaluation criteria include the significance, stability and discrimination of the features. The features that are important for the weld quality assessment are retained through screening, and redundant and invalid features are eliminated. Finally, the weld feature set extraction module combines the evaluated and screened features into a weld feature set, including the geometric morphology, texture features and structural features of the weld.

[0037] Furthermore, the weld feature set extraction module 12 in the platform provided in the application embodiment is also used for:

[0038] The multiple weld image information are preprocessed to obtain multiple weld image enhancement information; a historical weld image database is retrieved, the multiple weld image enhancement information are traversed for matching, and weld type information is generated; edge detection is performed on the multiple weld image enhancement information according to the weld type information combined with the background complexity of the multiple weld image enhancement information to generate weld edge parameters; grayscale co-occurrence matrix is ​​used for analysis to obtain multiple weld texture features; and evaluation is performed based on the weld edge parameters combined with the multiple weld texture features to generate the evaluation result.

[0039] In the embodiment of the present application, firstly, a plurality of weld image information is preprocessed to obtain a plurality of weld image enhancement information. Next, a feature matching algorithm, such as SIFT or SURF algorithm, is used to extract feature points from the weld image enhancement information. The feature point descriptors extracted by these algorithms characterize the local features in the image. Then, these feature point descriptors are compared with the feature points in the historical weld image database, and the best matching point pair is found by the nearest neighbor matching algorithm, and the weld type information is generated based on the matching results.

[0040] When edge detection is performed on multiple weld image enhancement information based on the weld type information combined with the background complexity of multiple weld image enhancement information, the Canny edge detection algorithm is used for edge detection. Specifically, the image is first smoothed by a Gaussian filter to remove noise; then the image gradient is calculated to identify the edge strength and direction in the image; then non-maximum suppression is performed to remove non-edge points and only retain local maxima; finally, the final edge is determined by double threshold detection and edge connection. The result of edge detection generates weld edge parameters and displays the contour and morphological characteristics of the weld.

[0041] Then, the texture analysis of the weld image is performed using the grayscale co-occurrence matrix. GLCM generates a co-occurrence matrix by counting the spatial co-occurrence relationship of pixel grayscale in the image. The specific steps include first calculating the grayscale co-occurrence matrix of the image in different directions and distances; then calculating the texture features based on the co-occurrence matrix, including contrast, entropy, energy, and uniformity. Contrast reflects the degree of change of grayscale values, entropy represents the complexity of image texture, energy reflects the stability of the image, and uniformity represents the uniformity of grayscale distribution. Through this process, multiple weld texture features are obtained, which describe the subtle structure and quality changes of the weld surface.

[0042] Based on the weld edge parameters and combined with multiple weld texture features, when comprehensively evaluating the weld image, the continuity, smoothness and consistency of the weld edge are first analyzed, and the morphological quality of the weld is evaluated by statistically analyzing parameters such as the edge length, curvature and number of breakpoints. Then, combined with the texture features, the fine structure and overall quality of the weld surface are evaluated. For example, the contrast and entropy reflect the uniformity and smoothness of the weld, and changes in energy and uniformity indicate whether there are defects or discontinuities in the weld. The final evaluation results are generated by comprehensively analyzing the weld edge parameters and texture features, providing a comprehensive evaluation of the weld quality.

[0043] Furthermore, the weld feature set extraction module 12 in the platform provided in the application embodiment is also used for:

[0044] The plurality of weld image information are filtered based on a filter to obtain a plurality of filtered image information; dynamic adjustment is performed based on the plurality of filtered image information in combination with histogram equalization to generate adjustment results; and the plurality of weld image enhancement information are obtained according to the adjustment results.

[0045] In the embodiment of the present application, when filtering multiple weld image information based on a filter, first select an appropriate filter, such as a Gaussian filter, to preprocess the weld image. The Gaussian filter generates a convolution kernel through a Gaussian function to convolve with the image, smooth the image, reduce high-frequency noise, and improve image quality. After applying the Gaussian filter, multiple filtered image information is generated.

[0046] Then, dynamic adjustment is performed based on multiple filtered image information combined with histogram equalization. Histogram equalization is an image enhancement technology that improves the contrast of an image by redistributing the grayscale values ​​of the image. Specifically, the grayscale histogram of the filtered image is first calculated, and then the histogram equalization algorithm is used to redistribute the grayscale values ​​so that the grayscale values ​​are evenly distributed throughout the range, thereby enhancing the contrast and detail of the image. After the histogram equalization process, the adjustment result is generated.

[0047] Finally, the adjustment results are recorded as multiple weld image enhancement information and output. The adjustment results improve the image contrast through histogram equalization, making the details of the weld more obvious and the edges clearer.

[0048] The labeling module 13 labels the plurality of weld image information based on the weld feature set to generate a plurality of image labeling information.

[0049] In the embodiment of the present application, the annotation module first receives a weld feature set, which contains key information in the weld image, such as the width, depth, shape, defect type and location of the weld. Using these feature sets, the annotation module matches and identifies each weld image, locates the weld by comparing the data in the feature set with the actual features in the image, and identifies the type of weld and possible defects.

[0050] After the weld and its features are identified, the annotation module begins to generate image annotation information. This information is presented in the form of graphic marks, such as rectangular boxes, line segments, arrows, etc. and text labels, which are used to clearly mark the location, size, type and any detected defects of the weld on the image. For example, for a crack defect in a weld, the annotation module draws a red line segment above it and attaches a text label of "crack".

[0051] The above process is repeated, and the annotation module generates corresponding annotation information for each weld image, thereby generating multiple image annotation information.

[0052] The weld quality detection module 14 constructs a weld quality detection model, synchronizes the multiple image annotation information to the weld quality detection model, and obtains a weld quality detection data set.

[0053] In the embodiment of the present application, the BP neural network is first trained with weld size data, weld surface defect data, and weld internal defect data to obtain a weld quality detection model. Next, multiple image annotation information is synchronized to the weld quality detection model to obtain a weld quality detection data set. The weld quality detection data set contains the weld quality assessment results after model analysis, including detailed information such as weld integrity, defect type, defect location, and severity.

[0054] Furthermore, the weld quality detection module 14 in the platform provided in the application embodiment is also used for:

[0055] Acquire weld size data, weld surface defect data and weld internal defect data; use the weld surface defect data and the weld internal defect data as input training data, and train the weld quality detection model based on the BP neural network; evaluate the accuracy of the output result of the weld quality detection model through the weld size data, and obtain an evaluation error parameter; iteratively train the weld quality detection model according to the evaluation error parameter, and when the number of consecutive iterations of the evaluation error parameter meets the preset requirements, it is deemed that the weld quality detection model has converged, and the construction of the weld quality detection model is completed.

[0056] In the embodiment of the present application, the weld size data, weld surface defect data and weld internal defect data are collected and acquired by the sensor and the image processing system. The weld size data includes geometric parameters such as the length, width and depth of the weld, while the weld surface defect data and internal defect data respectively include information such as surface cracks, holes and internal cracks, pores and slag inclusions.

[0057] After obtaining these data, the welding surface defect data and internal defect data are used as input training data to train the weld quality inspection model based on the BP neural network. The training process starts with the data preprocessing stage, normalizing the welding surface and internal defect data to ensure that the data is processed on the same scale, and performing data enhancement to improve the diversity of the data and the robustness of the model. Then initialize the weights and biases of the BP neural network model, set the number of layers of the network, the number of neurons in each layer, and the activation function and other parameters. Model training calculates the output results through forward propagation, and calculates the loss function, such as mean square error or cross entropy loss, by comparing it with the true defect label, and then adjusts the model parameters through the back propagation algorithm to minimize the loss function. This process is repeated in multiple iterations, gradually optimizing the model parameters until the training error converges to a smaller value.

[0058] After the initial model training is completed, the weld size data is input into the trained BP neural network model for evaluation. By comparing the model's prediction results with the actual weld size data, the accuracy and evaluation error parameters, such as mean square error, are calculated. The evaluation results are used to measure the model's prediction ability and accuracy.

[0059] According to the evaluation error parameters, the weld quality inspection model is iteratively trained. After the error is fed back to the BP neural network model, the model's training parameters, such as learning rate, weight and bias, are adjusted, and then training continues. This process continues, repeating forward propagation, back propagation and error feedback to continuously optimize the model parameters. The basis for convergence judgment during the training process is that the number of consecutive iterations of the evaluation error parameters meets the preset requirements, such as the error change is less than a certain threshold. At this time, the weld quality inspection model is considered to have converged and the training process is completed.

[0060] Through the above process, the construction of the weld quality inspection model is completed.

[0061] The retrospective verification module 15 performs retrospective verification on the weld quality inspection data set, updates the weld quality inspection data set according to the verification result, and transmits the updated result to the weld intelligent inspection platform.

[0062] In the embodiment of the present application, when performing retrospective verification, the length, width and depth of the weld are remeasured using a laser measuring instrument or an ultrasonic thickness gauge, and the measurement results are compared with the dimensional data in the test data set to ensure the accuracy of the dimensional data. Use a high-resolution camera and an optical microscope to conduct a detailed inspection of the weld surface to confirm the location, size and type of surface defects. Compare the inspection results with the surface defect data in the test data set to ensure the accuracy of the surface defect data. Use an ultrasonic flaw detector or an X-ray detector to inspect the inside of the weld to confirm the location, size and type of internal defects. Compare the inspection results with the internal defect data in the test data set to ensure the accuracy of the internal defect data.

[0063] According to the verification results, the retrospective verification module updates the weld quality inspection data set. If the inspection results in the data set are different from the verification results, the retrospective verification module updates the inspection data set, corrects the data of size, surface defects and internal defects, and finally transmits the updated results to the weld intelligent inspection platform.

[0064] Furthermore, the retrospective verification module 15 in the platform provided in the application embodiment is also used for:

[0065] Extract weld size detection data, welding surface defect detection data and welding internal defect detection data according to the weld quality detection data; perform back-testing on the target welding area according to the weld size detection data to generate a weld size back-testing verification result; perform back-testing on the target welding area according to the welding surface defect detection data to generate a welding surface defect back-testing verification result; perform back-testing on the target welding area according to the welding internal defect detection data to generate a welding internal defect back-testing verification result; perform abnormal correction on the weld quality detection data set according to the weld size back-testing verification result, the welding surface defect back-testing verification result and the welding internal defect back-testing verification result to generate the updated result.

[0066] In the embodiment of the present application, weld size detection data, welding surface defect detection data and welding internal defect detection data are first extracted from the weld quality detection data set. These data include the geometric dimensions of the weld, surface defects such as cracks and holes, and internal defects such as pores and slag inclusions.

[0067] Next, based on the weld size detection data, the target welding area is retrospectively verified to generate the weld size retrospective verification results. The specific method is to use a laser measuring instrument or an ultrasonic thickness gauge to re-measure the length, width and depth of the weld. The laser measuring instrument calculates the size of the weld using the reflection time of the laser beam, while the ultrasonic thickness gauge measures the thickness by the propagation time of the ultrasonic wave in the weld material. Through this process, the weld size retrospective verification results are obtained.

[0068] Next, based on the welding surface defect detection data, the target welding area is retrospectively verified to generate the welding surface defect retrospective verification results. The specific method is to use a high-resolution camera and an optical microscope to conduct a detailed inspection of the welding surface. The high-resolution camera captures a high-definition image of the weld surface, and the optical microscope magnifies the image to confirm the location, size and type of the surface defect. Through this process, the welding surface defect retrospective verification results are obtained.

[0069] Then, based on the welding internal defect detection data, the target welding area is retrospectively verified to generate the welding internal defect retrospective verification results. The specific method is to use an ultrasonic flaw detector or an X-ray detector to detect the inside of the weld. The ultrasonic flaw detector detects the location and size of internal defects through the reflection and transmission characteristics of ultrasonic waves, while the X-ray detector uses X-rays to penetrate the material to generate an image of the inside of the weld, showing the location and type of internal defects. Through this process, the welding internal defect retrospective verification results are obtained.

[0070] According to the weld size back-testing verification results, weld surface defect back-testing verification results, and weld internal defect back-testing verification results, the weld quality inspection data set is anomaly corrected to generate updated results. The specific steps include comparing each verification result with the corresponding data in the inspection data set to find out the differences and anomalies. Correct the erroneous data in the weld quality inspection data set based on the verification results. For example, if the laser measurement results show that the weld width is 2 mm wider than the original data, the weld width information in the data set is updated. The updated results are generated through the above process.

[0071] In the embodiments of the present application, in summary, the embodiments of the present application have at least the following technical effects:

[0072] The present application uses a sensor device group to perform real-time sensing acquisition of the target welding area to generate multiple weld image information; traverse multiple weld image information for evaluation, and extract a weld feature set based on the evaluation result; annotate the multiple weld image information based on the weld feature set to generate multiple image annotation information; build a weld quality detection model, synchronize multiple image annotation information to the weld quality detection model, and obtain a weld quality detection data set; perform retrospective verification on the weld quality detection data set, update the weld quality detection data set according to the verification result, and transmit the updated result to the weld intelligent detection platform. The present invention solves the technical problem that the prior art cannot realize the dynamic update of the weld quality detection data set, resulting in low detection accuracy, and achieves the technical effect of improving the weld detection accuracy through real-time sensing acquisition, weld feature extraction, image annotation, retrospective verification and dynamic update.

[0073] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the advantages and disadvantages of the embodiments. And the above-mentioned specific embodiments of this specification are described. The processes depicted in the accompanying drawings do not necessarily require the specific order and continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0074] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0075] This specification and the drawings are merely exemplary illustrations of the present application and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the present application. Obviously, those skilled in the art may make various modifications and variations to the present application without departing from the scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application intends to include these modifications and variations.

Claims

1. The weld seam intelligent detection platform based on artificial intelligence is characterized by: The platform includes: A sensor acquisition module, wherein the sensor acquisition module performs real-time sensor acquisition on the target welding area through a sensor device group to generate a plurality of weld image information; A weld feature set extraction module, wherein the weld feature set extraction module traverses the plurality of weld image information to perform evaluation and extracts a weld feature set according to the evaluation result; A labeling module, wherein the labeling module labels the plurality of weld image information based on the weld feature set to generate a plurality of image labeling information; A weld quality detection module, wherein the weld quality detection module constructs a weld quality detection model, synchronizes the plurality of image annotation information to the weld quality detection model, and obtains a weld quality detection data set; A retrospective verification module, wherein the retrospective verification module performs retrospective verification on the weld quality detection data set, updates the weld quality detection data set according to the verification result, and transmits the updated result to the weld intelligent detection platform; Wherein, the weld quality detection model platform includes: Obtain weld size data, weld surface defect data and weld internal defect data; Using the welding surface defect data and the welding internal defect data as input training data, and training the weld quality detection model based on a BP neural network; Performing an accuracy evaluation on the output result of the weld quality detection model using the weld size data to obtain an evaluation error parameter; The weld quality detection model is iteratively trained according to the evaluation error parameter. When the number of consecutive iterations of the evaluation error parameter meets the preset requirements, the weld quality detection model is deemed to have converged, and the construction of the weld quality detection model is completed.

2. The artificial intelligence-based weld intelligent detection platform according to claim 1, characterized in that: The sensor device group and platform include: Retrieving a historical weld sensor information set of a target welding area, wherein the historical weld sensor information set includes sensor light information, sensor coordinate information, and sensor range information; Determine a plurality of deployment points based on the sensing light information, the sensing coordinate information, and the sensing range information; Arranging a plurality of sensor devices based on the plurality of layout points, connecting the plurality of sensor devices according to a plurality of detection sensor requirement information of the target welding area, and generating a sensor topology structure; The sensing device group is generated based on the sensing topology.

3. The artificial intelligence-based weld intelligent detection platform according to claim 2, characterized in that: The sensor equipment group performs real-time sensing and acquisition of the target welding area to generate multiple weld image information. The platform includes: Starting the sensing device group to perform light balance sensing on the target welding area based on the sensing light information to obtain first sensing information; Performing position sensing on the target welding area based on the sensing coordinate information to obtain second sensing information; Performing multi-angle sensing on the target welding area based on the sensing range information to obtain third sensing information; Synchronize the first sensor information, the second sensor information, and the third sensor information to obtain sensor alignment data; The plurality of weld image information is extracted based on the sensor alignment data.

4. The artificial intelligence-based weld intelligent detection platform according to claim 3, characterized in that: Based on the sensing range information, multi-angle sensing is performed on the target welding area to obtain third sensing information. The platform includes: Dividing the sensing range information based on regional characteristics of the target welding area to obtain a plurality of sensing sub-areas; Traversing the multiple sensing sub-areas to perform multi-view matching and generate multiple sensing angles; Continuously sensing the weld according to the multiple sensing angles to generate multiple viewing angle image information; The multiple perspective images are fused according to a sensing time sequence to obtain the third sensing information.

5. The artificial intelligence-based weld intelligent detection platform according to claim 1, characterized in that: The plurality of weld image information are traversed for evaluation, and a weld feature set is extracted according to the evaluation result. The platform includes: Preprocessing the plurality of weld image information to obtain a plurality of weld image enhancement information; Retrieving a historical weld image database, traversing the plurality of weld image enhancement information for matching, and generating weld type information; Perform edge detection on the plurality of weld image enhancement information according to the weld type information and the background complexity of the plurality of weld image enhancement information to generate weld edge parameters; The gray-level co-occurrence matrix is ​​used for analysis to obtain multiple weld texture features; The evaluation result is generated by performing an evaluation based on the weld edge parameter in combination with the plurality of weld texture features.

6. The artificial intelligence-based weld intelligent detection platform according to claim 5, characterized in that: Preprocessing the plurality of weld image information to obtain a plurality of weld image enhancement information, the platform comprising: Filtering the plurality of weld image information based on a filter to obtain a plurality of filtered image information; Performing dynamic adjustment based on the plurality of filtered image information in combination with histogram equalization to generate an adjustment result; The plurality of weld image enhancement information are obtained according to the adjustment result.

7. The weld seam intelligent detection platform based on artificial intelligence according to claim 1, characterized in that: The weld quality inspection data set is retrospectively verified, and the weld quality inspection data set is updated according to the verification result. The platform includes: Extracting weld size detection data, welding surface defect detection data and welding internal defect detection data according to the weld quality detection data; Performing back-testing on the target welding area according to the weld size detection data to generate a weld size back-testing result; Performing retrospective verification on the target welding area according to the welding surface defect detection data to generate a welding surface defect retrospective verification result; Performing retrospective verification on the target welding area according to the welding internal defect detection data to generate a welding internal defect retrospective verification result; The weld quality inspection data set is anomaly corrected according to the weld size backtracking verification result, the weld surface defect backtracking verification result, and the weld internal defect backtracking verification result to generate the update result.

Images